DEVELOPING A NEURAL NETWORK-BASED METHOD FOR FASTER FACE RECOGNITION BY TRAINING & SIMULATION

摘要

Neural networks have been extensively studied for applications related to face recognition by training data sets extracted from facial images. In this paper, a feed-forward neural network-based upright frontal face detection method is developed. Local statistics of face images such as mean, standard deviation and kurtosis are used as datasets to be trained by feed forward neural networks. Each of the images is subdivided into several small blocks which are used to determine the datasets with two combinations of mean, standard deviation or kurtosis to train the feed-forward network. The extensive simulations are conducted using different images of same subjects from a standard face database to study and compare the face detection performance of the neural networks. It is found that neural network that uses the datasets of kurtosis gives the best performance among the two processes. Finally, the main focus is to improve the detection speed rather than performance.